Considerations for cellular quality control measures for the evolving diagnostic pathology landscape
By John Andersen, Clinical Associate Professor, Attending Pathologist, NYU Grossman School of Medicine
The convergence of digital pathology, generative artificial intelligence and big data analytics is changing the pathologic diagnostic landscape. This combination of precision pathology is driven by spatial biology characteristics to tailor personalized medicine for measuring the basic fundamental properties of cells within tissue. Digital pathology is rapidly evolving the way pathologists make their diagnoses whereby methods to ensure digital quality control need to be considered.
Digital pathology takes patient tissue on a microscope slide and digitizes it into an electronic image file. The major benefits are that the digital image can be viewed by anyone with a computer and at any location with proper imaging software. This technology is not new, but pathologic diagnosis utilizing a digital imaging modality has recently been accepted by the FDA and is compensated through Medicare. Deploying digital imaging platforms in major hospitals takes years in logistical management, but the upside is increased flexibility and diagnostic versatility.
Not only is digital imaging progressing the way pathologist’s render their diagnoses, the addition of generative artificial intelligence (AI) is refining the way qualitative and quantitative measures are assessed, improving test time and diagnostic precision. For example, with a trained artificial intelligence algorithm, the time it takes to measure the proliferative activity of a tumor is dramatically reduced from tens of minutes to seconds. The refined analysis can examine more precisely single cell counts within dynamic cellular communities and, thus, the immunologic regulatory landscape. This has an immediate impact on scenarios such as intraoperative consultation diagnosis, where a pathologist typically needs to impart a diagnosis that will dictate subsequent immediate surgical management. Computational discrimination would allow for quality control in the single-cell setting.
The cell line quality control improvements can be further escalated in the digital pathology world. Single cell staining is cleaner and would be less prone to technical errors.
The rapidly evolving oncologic treatment landscape has led to a dramatic increase in the utilization of therapeutic immunohistologic markers such PDL-1, HER-2, BRAF, Ki-67, p16 and recently FOLR1. These markers are FDA-approved in some capacity as an adjunct for therapeutic decision-making. For example, Herceptin (trastuzumab) has been recently approved as a line of treatment for any resistant metastatic carcinoma. Now, HER2 immunohistochemical testing is requested by oncologists for nearly all metastatic tissue and primary diagnosis tissues.
Pathologists interpret these immunohistochemical markers through a scoring mechanism that includes either a positive or negative or binary range, small range (e.g., 2-5%), wide range (quartile), or semi-quantitatively, such as the amount of cell membrane staining. Unfortunately, interobserver and intraobserver variability in quantitative assessment is higher than most non-medical professionals would assume. Many groups claim that their group’s consensus or experience offsets these variabilities, but variability amongst even experienced professionals is significant.
This is why robust, readily available quality control measures are needed to ensure that factors outside the control of the pathologist are minimized and that standardized controls are used in conjunction with patient tissue. The biochemical mechanisms that immunohistochemistry utilizes are sensitive to daily variation so that a control, often patient tissue that has been designated as a standard control, is placed on the same slide to ensure that expected positives are positive and expected negatives are negative. These controls act as a quality control mechanism that ensures that every immunohistochemical stain that is performed is achieving optimal and desired results. However, variability in results is typical and some degree of experience in interpretation is needed to accurately assess these subtle variations.
Assessment of the amount of protein and location of expression are necessary for a thorough interpretation of immunohistochemical tests. Therapeutic marker evaluation requires an experienced pathologist, often with subspecialty expertise, for accurate interpretation. For example, PDL-1 protein, which, when expressed, indicates the likelihood that a therapy will be effective. However, assessing its expression can be complex, involving a review of cellular compartments and surrounding tissue inflammatory cells. Furthermore, tumors with heterogenous composition can vary in their expression of PDL-1, complicating assessment.
Further diversifying the utility of quality control mechanisms is the assessment for dysplasia, which is a precursor lesion to malignancy. With precision medicine increasing diagnostic accuracy and expanding tumor diagnostic versatility, the movement towards catching malignancy before it occurs is increasing. An example is the assessment of gastric and oropharyngeal squamous cell dysplasia. TP53, which is the most commonly mutated gene in cancer and is involved in many cancers, is commonly assessed using immunohistochemistry as an adjunct for identifying its mutation status. Increased p53 expression indicates that dysplasia is occurring and that the lesional tissue being examined represents a true precursor to cancer.
New quality control methods with increased availability are needed to accommodate the evolving pathology landscape. One particularly intriguing method is the use of cell lines that are screened for desired expression profiles that match tumor type expression. Instead of using patient tissue that is limited to either positive or negative binary ranges and limited to major medical centers that can store tissue, individual cell lines can be propagated endlessly and aligned to fit desired wide-range expression profiles. For example, p53 protein expression which has three different expression profiles; completely negative or loss of protein, normal expression which is typically weak and scattered expression and strong diffuse expression. Creating a cell line quality control panel that demonstrates these three expression panels would increase quality measures.
Furthermore, using this cell line panel could improve antibody optimization, increase clarity, and produce a more representative full range of expression patterns. It could act as an adjunct to cytologic preparations, which are comprised of cells and not tissue. Additionally, it has the potential to improve diagnostic accuracy for the less experienced pathologist especially in the setting of precursor lesions. Cell lines have increased utility and versatility, which decreases the reliance on patient tumor tissue.
The cell line quality control improvements can be further escalated in the digital pathology world. Single cell staining is cleaner and would be less prone to technical errors. The decrease in technical errors would allow for improved optical resolution and precise digital assessment. Increased diagnostic accuracy and improved digital imaging would lead to more refined big data analytics, increasing the potential for better outcomes with future translational medicine initiatives.